Buy and sell patterns, which are frequently detected using technical indicators, are the main forces behind trading dynamics in the stock market. A key instrument for assessing overbought or oversold conditions is the Relative Strength Index (RSI), a momentum oscillator that gauges the magnitude of recent price fluctuations. This study uses the prediction outputs of a unique neural network (NN) framework to classify market conditions into buy, neutral, or sell states. Particularly, this paper presents a three-layer metaheuristic fuzzy adaptive weights and structure determination NN, called FCBW, to address the inherent drawbacks of conventional back-propagation NNs, including slow training speeds and vulnerability to local minima. The FCBW model, which is created especially for multiclass classification, makes use of fuzzy logic, a bio-inspired algorithm, and a direct weight determination technique to improve architectural stability. The FCBW model greatly outperforms well-known machine learning benchmarks, according to experimental results on two of the most active equities in the US market. The suggested model provides improved robustness and accuracy for algorithmic trading environments by efficiently categorizing buy, sell, and neutral states based on anticipated market movements.